英文翻译Intelligence Control and its ApplicationZOU En1,2,LIN Yi-qin3,ZHANG Tai-Shan1Abstract: The author introduces some basic concepts about intelligence control in this paper, which includes fuzzy control, adaptive fuzzy-neural control, expert fuzzy system and artificial neural networks and so on, and features of fuzzy theory and artificial neural network are briefly analyzed. Finally, the author combines artificial neural network techniques with PID control, the neural network PID control method is applied to the temperature control system, seeing from the output curve, the present method has many advantages of small overshoot, short setting-time, and an excellent control result has provided for the system.Key words: intelligence control; neural network; temperature control system.It is well known that the emergence of intelligent control has made it conducive to integrate fuzz logic control and artificial neural networks and expert system for the development of control systems. The ability of these systems to complex non-linear function of many variables through a learning control driven by an error signal is particularly attractive. On the other hand, the intelligent control has found various applications in industrial produce as well as household applications. For example, for some complex or ill-defined systems that are not easily controlled by conventional schemes, the advanced control can provide a feasible alternative to get the approximate and qualitative data of human experts knowledge. As developing industrial control, the flourishing field of intelligent control technology has already provided some significant applications in the new control technology.Some basic concepts about intelligent control have been attempted to set out in this paper and also uses an example to show the application of intelligent control in the temperature control system.1fuzzy controlThe development of fuzzy theory comes from the inability to describe some physical phenomena with the exact mathematical or conventional model. So, fuzzy theory is a powerful tool in the exploration of complex problems, because of it has ability to determine outputs for a given set of inputs without using a conventional or mathematical model. it is a no model controller The basic motivation of fuzzy theory is that complex problems become simplicity [1-3].Fuzzy subset A of an universe of discourse U is characterized by a membership functionμ(χ):χ∈→[0,1],representing the grade of membership of χin A .Fuzzy theory owes a great Adeal to human language, it is a language controller; each word or linguist term in a natural language can be viewed as alabel for a fuzzy subset A of a universe of discourse U. This language labels describing words,phrases and sentences to subsets of U. A fuzzy linguistic variable is a variable whose values are linguistic terms used as labels of fuzzy sets. For example, the fuzzy subset labels high, medium, and low can be regarded as values of the fuzzy variable.2The adaptive fuzzy-neural controlThe adaptive fuzzy-neural control commonly consists of two multi-layered neural network models configured in the architecture. The first neural network is a plant emulator and the second neural networkis used as a compensator to improve the performance of the basic fuzzy logic controller. The development of this system consists of three phases. The first phase is developing a basic fuzzy logic controller for the plant. The second phase involves training a neural network model the forward dynamics of the plant to be controlled. The training of this neural network can be done off-line as well as on-line depending on the type of plant. The third phase involves on-line learning of the neural fuzzy compensator. The performance error, which is the error between the desired output and the actual output, is back propagated through the neural plant emulator to adapt the weights of the neural fuzzy compensator on-line. The performance further improved on-line by back propagation propagator of the error between the neural plant emulator and the actual plant output.3Expert Fuzzy systemAn expert system was a program system that it had a lot of expert knowledge and experience; it had been developed with the expert's knowledge and previous experimental data. In order to express the expert's knowledge graphically, a knowledge network was implemented for illustration of causal relationships[4]; fuzzy membership functions were used for linguistic representations. when the expert system detects faults, it starts to search for the original causes of them by using backward and forward chaining methods. Then, according to the original causes modify control strategy; for every operation it calculates values, which represents degree of certainty for the operation. If the value of an original cause is above the predefined limit, the expert system decides to execute the operation. When the value is below the predefined limit and if the operation is irreversible, the expert system gives a message to the operator to get his decision. If the operation is reversible, the expert system decides to use the operation. Three different control strategies were selected in study, such as: “Set point”,“Fuzzy Answer”and “Advice”to the operator. When the expert system decides to use “Set point”it sends a new value of process variable to high level control system. The “Advice”will be sent to the process operator in order to perform manual operation. The “Fuzzy Answer” consisted of three parts, a process variable, the respective predefined fuzzy membership function, and the degree of certainly of the discovered original cause, for each discovered fault a "FuzzyAnswer” is created.4Artificial neuralnetworksArtificial neural networks try to imitate the biological brain neural network into mathematical model. The brain is large-scale information's processing system connecting about 1010 neurons. The artificial neural network connects many linear or nonlinear neuron models and processes information in a parallel-distributed manner. While the computation speed of conventional computers is limited by the servant computation scheme with pre-assigned algorithm, the neural networks can perform computations at much higher speed [5,6].In addition, the neural network has many interesting and attractive features, such as, large parallel processing, fault tolerant, adaptive learning, and self-organization capabilities.An artificial neural network is a collection of neural units gathered in different layers. A typical multiplayer neural network is shown as follows.Multilayer networks can implement arbitrary complex input-output mappings. The output of a neuron i in the kth layer is as follows:)(11j k iN i k ij k i xw f y θ-=-=∑ , k= 1, 2,…m . (1) Where y i k is the output of the ith neuron in the kth layer, and w y k is the connection weight fromthe ith neuron of the (k-i)th layer to the jth neuron of the kth layers, m is total numbers of layers, X i k-1 is the activation of the ith neuron of the(k-1) layer, θj is the threshold value of the jth neuron. The .f (·) functions represents the activation rule of the neuron which is normally a step, ramp, linear or sigmoid function. In a competitive neural network, each neuron i in the kth layer is in competition with the other neurons of the same layers. In order to learn the weights of neural network, error back propagation algorithm can be used. This algorithm uses gradient search technique for minimizing the error function. Recently, the neural network usual is combined other some controllers by using the control system, such as the neural network PID controller, neural network fuzzy controller …,facts illustrate that combinative control effects are better than individual.5.Simulation examplesThe realization of neural network control in a temperature control system of boiler is shown in Fib. 2Controller is neural network PID control schema , The structure figure of temperature control system is shown in Fig. 3.where the neural network PID regulator is a two-layer network control system is shown in Fib. 4.where x 0=1,x 1=e(t),x 2=∑=tn n e 0)(,x 3=△e(t)-e(t-1) the object function can be defined as follows:∑=-==t n n n y r J 0)(21 , ( 2 ) )()(∙=f t u , xe xf -+=11)(Algorithm is on the gradient method, which is called the back-propagation method, the backward error signal between two layers is expressed as)()]()([)('x f t u t J t -=δ , (3)where f(x) is a derivative of f with respect to x.the connection weight is as follows:)()()()()1(n w t u t t n w iu iu iu ∆+∙∙∙-=+∆αδη , (4)WhereLearning rate )]1(exp[)(0--∙=t J t ηη , (5)Intensive factor )]1(exp[)(0--=t J t αα , (6)Order η0=0.3 ,α0=0.95 .W 。